Pingchuan Ma

Hi @Winfredy, we use the [RetinaFace](https://openaccess.thecvf.com/content_CVPR_2020/papers/Deng_RetinaFace_Single-Shot_Multi-Level_Face_Localisation_in_the_Wild_CVPR_2020_paper.pdf) as the face detector and use the 2D facial landmark detector based on [FAN](http://openaccess.thecvf.com/content_ICCV_2017/papers/Bulat_How_Far_Are_ICCV_2017_paper.pdf).

Hi, we use noisy data augmentation. It's only added in training. The model could be evaluated in different noisy scenarios. Regarding the adding method, it’s similar to [this](https://github.com/mpc001/Lipreading_using_Temporal_Convolutional_Networks/blob/fdc0f1a6d5da93246f9aa6e339bcd3f0d9703c0a/lipreading/preprocess.py#L150).

A similar issue was raised in #16. Can you please check that you have correctly specified the file path for Windows operating system?

Duplicate of #14

Hi, the landmark can be extracted via face tracker. You could check dlib (#1), or face alignment library (#10). We also provide a RetinaFace tracker but it's at [another repository](https://github.com/mpc001/Visual_Speech_Recognition_for_Multiple_Languages#how-to-prepare-models-and-landmarks).

Hi, Please check the attached babble noisy file [babble noise](https://drive.google.com/file/d/1uNo0PaGI-jyNTM-zR_OrNWrppaM-N6SS/view?usp=sharing). You could add a random clip of the noise to the audio and get noisy audio with a specified SNR...

Duplicate of #23

Hi, Thanks for your interests! It's better to use 83.3, which is the update. Kind regards, Pingchuan

Hi, could you check if you've installed `ffmpeg`?

Hello, that's a weird result on the test set. When testing a model without word boundaries, have you encountered a similar issue? By the way, when testing a model that...